msi.c revision 277bc33bc2479707e88b0b2ae6fe56e8e4aabe81
1/*
2 * File:	msi.c
3 * Purpose:	PCI Message Signaled Interrupt (MSI)
4 *
5 * Copyright (C) 2003-2004 Intel
6 * Copyright (C) Tom Long Nguyen (tom.l.nguyen@intel.com)
7 */
8
9#include <linux/err.h>
10#include <linux/mm.h>
11#include <linux/irq.h>
12#include <linux/interrupt.h>
13#include <linux/init.h>
14#include <linux/ioport.h>
15#include <linux/smp_lock.h>
16#include <linux/pci.h>
17#include <linux/proc_fs.h>
18
19#include <asm/errno.h>
20#include <asm/io.h>
21#include <asm/smp.h>
22
23#include "pci.h"
24#include "msi.h"
25
26static DEFINE_SPINLOCK(msi_lock);
27static struct msi_desc* msi_desc[NR_IRQS] = { [0 ... NR_IRQS-1] = NULL };
28static kmem_cache_t* msi_cachep;
29
30static int pci_msi_enable = 1;
31
32static struct msi_ops *msi_ops;
33
34int
35msi_register(struct msi_ops *ops)
36{
37	msi_ops = ops;
38	return 0;
39}
40
41static int msi_cache_init(void)
42{
43	msi_cachep = kmem_cache_create("msi_cache", sizeof(struct msi_desc),
44					0, SLAB_HWCACHE_ALIGN, NULL, NULL);
45	if (!msi_cachep)
46		return -ENOMEM;
47
48	return 0;
49}
50
51static void msi_set_mask_bit(unsigned int irq, int flag)
52{
53	struct msi_desc *entry;
54
55	entry = msi_desc[irq];
56	BUG_ON(!entry || !entry->dev);
57	switch (entry->msi_attrib.type) {
58	case PCI_CAP_ID_MSI:
59		if (entry->msi_attrib.maskbit) {
60			int		pos;
61			u32		mask_bits;
62
63			pos = (long)entry->mask_base;
64			pci_read_config_dword(entry->dev, pos, &mask_bits);
65			mask_bits &= ~(1);
66			mask_bits |= flag;
67			pci_write_config_dword(entry->dev, pos, mask_bits);
68		}
69		break;
70	case PCI_CAP_ID_MSIX:
71	{
72		int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
73			PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
74		writel(flag, entry->mask_base + offset);
75		break;
76	}
77	default:
78		BUG();
79		break;
80	}
81}
82
83static void read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
84{
85	switch(entry->msi_attrib.type) {
86	case PCI_CAP_ID_MSI:
87	{
88		struct pci_dev *dev = entry->dev;
89		int pos = entry->msi_attrib.pos;
90		u16 data;
91
92		pci_read_config_dword(dev, msi_lower_address_reg(pos),
93					&msg->address_lo);
94		if (entry->msi_attrib.is_64) {
95			pci_read_config_dword(dev, msi_upper_address_reg(pos),
96						&msg->address_hi);
97			pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
98		} else {
99			msg->address_hi = 0;
100			pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
101		}
102		msg->data = data;
103		break;
104	}
105	case PCI_CAP_ID_MSIX:
106	{
107		void __iomem *base;
108		base = entry->mask_base +
109			entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
110
111		msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
112		msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
113		msg->data = readl(base + PCI_MSIX_ENTRY_DATA_OFFSET);
114 		break;
115 	}
116 	default:
117		BUG();
118	}
119}
120
121static void write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
122{
123	switch (entry->msi_attrib.type) {
124	case PCI_CAP_ID_MSI:
125	{
126		struct pci_dev *dev = entry->dev;
127		int pos = entry->msi_attrib.pos;
128
129		pci_write_config_dword(dev, msi_lower_address_reg(pos),
130					msg->address_lo);
131		if (entry->msi_attrib.is_64) {
132			pci_write_config_dword(dev, msi_upper_address_reg(pos),
133						msg->address_hi);
134			pci_write_config_word(dev, msi_data_reg(pos, 1),
135						msg->data);
136		} else {
137			pci_write_config_word(dev, msi_data_reg(pos, 0),
138						msg->data);
139		}
140		break;
141	}
142	case PCI_CAP_ID_MSIX:
143	{
144		void __iomem *base;
145		base = entry->mask_base +
146			entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
147
148		writel(msg->address_lo,
149			base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
150		writel(msg->address_hi,
151			base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
152		writel(msg->data, base + PCI_MSIX_ENTRY_DATA_OFFSET);
153		break;
154	}
155	default:
156		BUG();
157	}
158}
159
160#ifdef CONFIG_SMP
161static void set_msi_affinity(unsigned int irq, cpumask_t cpu_mask)
162{
163	struct msi_desc *entry;
164	struct msi_msg msg;
165
166	entry = msi_desc[irq];
167	if (!entry || !entry->dev)
168		return;
169
170	read_msi_msg(entry, &msg);
171	msi_ops->target(irq, cpu_mask, &msg);
172	write_msi_msg(entry, &msg);
173	set_native_irq_info(irq, cpu_mask);
174}
175#else
176#define set_msi_affinity NULL
177#endif /* CONFIG_SMP */
178
179static void mask_MSI_irq(unsigned int irq)
180{
181	msi_set_mask_bit(irq, 1);
182}
183
184static void unmask_MSI_irq(unsigned int irq)
185{
186	msi_set_mask_bit(irq, 0);
187}
188
189static void ack_msi_irq(unsigned int irq)
190{
191	move_native_irq(irq);
192	ack_APIC_irq();
193}
194
195/*
196 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
197 * which implement the MSI or MSI-X Capability Structure.
198 */
199static struct irq_chip msi_chip = {
200	.name		= "PCI-MSI",
201	.unmask		= unmask_MSI_irq,
202	.mask		= mask_MSI_irq,
203	.ack		= ack_msi_irq,
204	.set_affinity	= set_msi_affinity
205};
206
207static int msi_free_irq(struct pci_dev* dev, int irq);
208static int msi_init(void)
209{
210	static int status = -ENOMEM;
211
212	if (!status)
213		return status;
214
215	if (pci_msi_quirk) {
216		pci_msi_enable = 0;
217		printk(KERN_WARNING "PCI: MSI quirk detected. MSI disabled.\n");
218		status = -EINVAL;
219		return status;
220	}
221
222	status = msi_arch_init();
223	if (status < 0) {
224		pci_msi_enable = 0;
225		printk(KERN_WARNING
226		       "PCI: MSI arch init failed.  MSI disabled.\n");
227		return status;
228	}
229
230	if (! msi_ops) {
231		pci_msi_enable = 0;
232		printk(KERN_WARNING
233		       "PCI: MSI ops not registered. MSI disabled.\n");
234		status = -EINVAL;
235		return status;
236	}
237
238	status = msi_cache_init();
239	if (status < 0) {
240		pci_msi_enable = 0;
241		printk(KERN_WARNING "PCI: MSI cache init failed\n");
242		return status;
243	}
244
245	return status;
246}
247
248static struct msi_desc* alloc_msi_entry(void)
249{
250	struct msi_desc *entry;
251
252	entry = kmem_cache_zalloc(msi_cachep, GFP_KERNEL);
253	if (!entry)
254		return NULL;
255
256	entry->link.tail = entry->link.head = 0;	/* single message */
257	entry->dev = NULL;
258
259	return entry;
260}
261
262static void attach_msi_entry(struct msi_desc *entry, int irq)
263{
264	unsigned long flags;
265
266	spin_lock_irqsave(&msi_lock, flags);
267	msi_desc[irq] = entry;
268	spin_unlock_irqrestore(&msi_lock, flags);
269}
270
271static int create_msi_irq(struct irq_chip *chip)
272{
273	struct msi_desc *entry;
274	int irq;
275
276	entry = alloc_msi_entry();
277	if (!entry)
278		return -ENOMEM;
279
280	irq = create_irq();
281	if (irq < 0) {
282		kmem_cache_free(msi_cachep, entry);
283		return -EBUSY;
284	}
285
286	set_irq_chip_and_handler(irq, chip, handle_edge_irq);
287	set_irq_data(irq, entry);
288
289	return irq;
290}
291
292static void destroy_msi_irq(unsigned int irq)
293{
294	struct msi_desc *entry;
295
296	entry = get_irq_data(irq);
297	set_irq_chip(irq, NULL);
298	set_irq_data(irq, NULL);
299	destroy_irq(irq);
300	kmem_cache_free(msi_cachep, entry);
301}
302
303static void enable_msi_mode(struct pci_dev *dev, int pos, int type)
304{
305	u16 control;
306
307	pci_read_config_word(dev, msi_control_reg(pos), &control);
308	if (type == PCI_CAP_ID_MSI) {
309		/* Set enabled bits to single MSI & enable MSI_enable bit */
310		msi_enable(control, 1);
311		pci_write_config_word(dev, msi_control_reg(pos), control);
312		dev->msi_enabled = 1;
313	} else {
314		msix_enable(control);
315		pci_write_config_word(dev, msi_control_reg(pos), control);
316		dev->msix_enabled = 1;
317	}
318    	if (pci_find_capability(dev, PCI_CAP_ID_EXP)) {
319		/* PCI Express Endpoint device detected */
320		pci_intx(dev, 0);  /* disable intx */
321	}
322}
323
324void disable_msi_mode(struct pci_dev *dev, int pos, int type)
325{
326	u16 control;
327
328	pci_read_config_word(dev, msi_control_reg(pos), &control);
329	if (type == PCI_CAP_ID_MSI) {
330		/* Set enabled bits to single MSI & enable MSI_enable bit */
331		msi_disable(control);
332		pci_write_config_word(dev, msi_control_reg(pos), control);
333		dev->msi_enabled = 0;
334	} else {
335		msix_disable(control);
336		pci_write_config_word(dev, msi_control_reg(pos), control);
337		dev->msix_enabled = 0;
338	}
339    	if (pci_find_capability(dev, PCI_CAP_ID_EXP)) {
340		/* PCI Express Endpoint device detected */
341		pci_intx(dev, 1);  /* enable intx */
342	}
343}
344
345static int msi_lookup_irq(struct pci_dev *dev, int type)
346{
347	int irq;
348	unsigned long flags;
349
350	spin_lock_irqsave(&msi_lock, flags);
351	for (irq = 0; irq < NR_IRQS; irq++) {
352		if (!msi_desc[irq] || msi_desc[irq]->dev != dev ||
353			msi_desc[irq]->msi_attrib.type != type ||
354			msi_desc[irq]->msi_attrib.default_irq != dev->irq)
355			continue;
356		spin_unlock_irqrestore(&msi_lock, flags);
357		/* This pre-assigned MSI irq for this device
358		   already exits. Override dev->irq with this irq */
359		dev->irq = irq;
360		return 0;
361	}
362	spin_unlock_irqrestore(&msi_lock, flags);
363
364	return -EACCES;
365}
366
367void pci_scan_msi_device(struct pci_dev *dev)
368{
369	if (!dev)
370		return;
371}
372
373#ifdef CONFIG_PM
374int pci_save_msi_state(struct pci_dev *dev)
375{
376	int pos, i = 0;
377	u16 control;
378	struct pci_cap_saved_state *save_state;
379	u32 *cap;
380
381	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
382	if (pos <= 0 || dev->no_msi)
383		return 0;
384
385	pci_read_config_word(dev, msi_control_reg(pos), &control);
386	if (!(control & PCI_MSI_FLAGS_ENABLE))
387		return 0;
388
389	save_state = kzalloc(sizeof(struct pci_cap_saved_state) + sizeof(u32) * 5,
390		GFP_KERNEL);
391	if (!save_state) {
392		printk(KERN_ERR "Out of memory in pci_save_msi_state\n");
393		return -ENOMEM;
394	}
395	cap = &save_state->data[0];
396
397	pci_read_config_dword(dev, pos, &cap[i++]);
398	control = cap[0] >> 16;
399	pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO, &cap[i++]);
400	if (control & PCI_MSI_FLAGS_64BIT) {
401		pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI, &cap[i++]);
402		pci_read_config_dword(dev, pos + PCI_MSI_DATA_64, &cap[i++]);
403	} else
404		pci_read_config_dword(dev, pos + PCI_MSI_DATA_32, &cap[i++]);
405	if (control & PCI_MSI_FLAGS_MASKBIT)
406		pci_read_config_dword(dev, pos + PCI_MSI_MASK_BIT, &cap[i++]);
407	save_state->cap_nr = PCI_CAP_ID_MSI;
408	pci_add_saved_cap(dev, save_state);
409	return 0;
410}
411
412void pci_restore_msi_state(struct pci_dev *dev)
413{
414	int i = 0, pos;
415	u16 control;
416	struct pci_cap_saved_state *save_state;
417	u32 *cap;
418
419	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_MSI);
420	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
421	if (!save_state || pos <= 0)
422		return;
423	cap = &save_state->data[0];
424
425	control = cap[i++] >> 16;
426	pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO, cap[i++]);
427	if (control & PCI_MSI_FLAGS_64BIT) {
428		pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI, cap[i++]);
429		pci_write_config_dword(dev, pos + PCI_MSI_DATA_64, cap[i++]);
430	} else
431		pci_write_config_dword(dev, pos + PCI_MSI_DATA_32, cap[i++]);
432	if (control & PCI_MSI_FLAGS_MASKBIT)
433		pci_write_config_dword(dev, pos + PCI_MSI_MASK_BIT, cap[i++]);
434	pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
435	enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
436	pci_remove_saved_cap(save_state);
437	kfree(save_state);
438}
439
440int pci_save_msix_state(struct pci_dev *dev)
441{
442	int pos;
443	int temp;
444	int irq, head, tail = 0;
445	u16 control;
446	struct pci_cap_saved_state *save_state;
447
448	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
449	if (pos <= 0 || dev->no_msi)
450		return 0;
451
452	/* save the capability */
453	pci_read_config_word(dev, msi_control_reg(pos), &control);
454	if (!(control & PCI_MSIX_FLAGS_ENABLE))
455		return 0;
456	save_state = kzalloc(sizeof(struct pci_cap_saved_state) + sizeof(u16),
457		GFP_KERNEL);
458	if (!save_state) {
459		printk(KERN_ERR "Out of memory in pci_save_msix_state\n");
460		return -ENOMEM;
461	}
462	*((u16 *)&save_state->data[0]) = control;
463
464	/* save the table */
465	temp = dev->irq;
466	if (msi_lookup_irq(dev, PCI_CAP_ID_MSIX)) {
467		kfree(save_state);
468		return -EINVAL;
469	}
470
471	irq = head = dev->irq;
472	while (head != tail) {
473		struct msi_desc *entry;
474
475		entry = msi_desc[irq];
476		read_msi_msg(entry, &entry->msg_save);
477
478		tail = msi_desc[irq]->link.tail;
479		irq = tail;
480	}
481	dev->irq = temp;
482
483	save_state->cap_nr = PCI_CAP_ID_MSIX;
484	pci_add_saved_cap(dev, save_state);
485	return 0;
486}
487
488void pci_restore_msix_state(struct pci_dev *dev)
489{
490	u16 save;
491	int pos;
492	int irq, head, tail = 0;
493	struct msi_desc *entry;
494	int temp;
495	struct pci_cap_saved_state *save_state;
496
497	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_MSIX);
498	if (!save_state)
499		return;
500	save = *((u16 *)&save_state->data[0]);
501	pci_remove_saved_cap(save_state);
502	kfree(save_state);
503
504	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
505	if (pos <= 0)
506		return;
507
508	/* route the table */
509	temp = dev->irq;
510	if (msi_lookup_irq(dev, PCI_CAP_ID_MSIX))
511		return;
512	irq = head = dev->irq;
513	while (head != tail) {
514		entry = msi_desc[irq];
515		write_msi_msg(entry, &entry->msg_save);
516
517		tail = msi_desc[irq]->link.tail;
518		irq = tail;
519	}
520	dev->irq = temp;
521
522	pci_write_config_word(dev, msi_control_reg(pos), save);
523	enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
524}
525#endif
526
527static int msi_register_init(struct pci_dev *dev, struct msi_desc *entry)
528{
529	int status;
530	struct msi_msg msg;
531	int pos;
532	u16 control;
533
534	pos = entry->msi_attrib.pos;
535	pci_read_config_word(dev, msi_control_reg(pos), &control);
536
537	/* Configure MSI capability structure */
538	status = msi_ops->setup(dev, dev->irq, &msg);
539	if (status < 0)
540		return status;
541
542	write_msi_msg(entry, &msg);
543	if (entry->msi_attrib.maskbit) {
544		unsigned int maskbits, temp;
545		/* All MSIs are unmasked by default, Mask them all */
546		pci_read_config_dword(dev,
547			msi_mask_bits_reg(pos, is_64bit_address(control)),
548			&maskbits);
549		temp = (1 << multi_msi_capable(control));
550		temp = ((temp - 1) & ~temp);
551		maskbits |= temp;
552		pci_write_config_dword(dev,
553			msi_mask_bits_reg(pos, is_64bit_address(control)),
554			maskbits);
555	}
556
557	return 0;
558}
559
560/**
561 * msi_capability_init - configure device's MSI capability structure
562 * @dev: pointer to the pci_dev data structure of MSI device function
563 *
564 * Setup the MSI capability structure of device function with a single
565 * MSI irq, regardless of device function is capable of handling
566 * multiple messages. A return of zero indicates the successful setup
567 * of an entry zero with the new MSI irq or non-zero for otherwise.
568 **/
569static int msi_capability_init(struct pci_dev *dev)
570{
571	int status;
572	struct msi_desc *entry;
573	int pos, irq;
574	u16 control;
575
576   	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
577	pci_read_config_word(dev, msi_control_reg(pos), &control);
578	/* MSI Entry Initialization */
579	irq = create_msi_irq(&msi_chip);
580	if (irq < 0)
581		return irq;
582
583	entry = get_irq_data(irq);
584	entry->link.head = irq;
585	entry->link.tail = irq;
586	entry->msi_attrib.type = PCI_CAP_ID_MSI;
587	entry->msi_attrib.is_64 = is_64bit_address(control);
588	entry->msi_attrib.entry_nr = 0;
589	entry->msi_attrib.maskbit = is_mask_bit_support(control);
590	entry->msi_attrib.default_irq = dev->irq;	/* Save IOAPIC IRQ */
591	entry->msi_attrib.pos = pos;
592	dev->irq = irq;
593	entry->dev = dev;
594	if (is_mask_bit_support(control)) {
595		entry->mask_base = (void __iomem *)(long)msi_mask_bits_reg(pos,
596				is_64bit_address(control));
597	}
598	/* Configure MSI capability structure */
599	status = msi_register_init(dev, entry);
600	if (status != 0) {
601		dev->irq = entry->msi_attrib.default_irq;
602		destroy_msi_irq(irq);
603		return status;
604	}
605
606	attach_msi_entry(entry, irq);
607	/* Set MSI enabled bits	 */
608	enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
609
610	return 0;
611}
612
613/**
614 * msix_capability_init - configure device's MSI-X capability
615 * @dev: pointer to the pci_dev data structure of MSI-X device function
616 * @entries: pointer to an array of struct msix_entry entries
617 * @nvec: number of @entries
618 *
619 * Setup the MSI-X capability structure of device function with a
620 * single MSI-X irq. A return of zero indicates the successful setup of
621 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
622 **/
623static int msix_capability_init(struct pci_dev *dev,
624				struct msix_entry *entries, int nvec)
625{
626	struct msi_desc *head = NULL, *tail = NULL, *entry = NULL;
627	struct msi_msg msg;
628	int status;
629	int irq, pos, i, j, nr_entries, temp = 0;
630	unsigned long phys_addr;
631	u32 table_offset;
632 	u16 control;
633	u8 bir;
634	void __iomem *base;
635
636   	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
637	/* Request & Map MSI-X table region */
638 	pci_read_config_word(dev, msi_control_reg(pos), &control);
639	nr_entries = multi_msix_capable(control);
640
641 	pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
642	bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
643	table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
644	phys_addr = pci_resource_start (dev, bir) + table_offset;
645	base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
646	if (base == NULL)
647		return -ENOMEM;
648
649	/* MSI-X Table Initialization */
650	for (i = 0; i < nvec; i++) {
651		irq = create_msi_irq(&msi_chip);
652		if (irq < 0)
653			break;
654
655		entry = get_irq_data(irq);
656 		j = entries[i].entry;
657 		entries[i].vector = irq;
658		entry->msi_attrib.type = PCI_CAP_ID_MSIX;
659		entry->msi_attrib.is_64 = 1;
660		entry->msi_attrib.entry_nr = j;
661		entry->msi_attrib.maskbit = 1;
662		entry->msi_attrib.default_irq = dev->irq;
663		entry->msi_attrib.pos = pos;
664		entry->dev = dev;
665		entry->mask_base = base;
666		if (!head) {
667			entry->link.head = irq;
668			entry->link.tail = irq;
669			head = entry;
670		} else {
671			entry->link.head = temp;
672			entry->link.tail = tail->link.tail;
673			tail->link.tail = irq;
674			head->link.head = irq;
675		}
676		temp = irq;
677		tail = entry;
678		/* Configure MSI-X capability structure */
679		status = msi_ops->setup(dev, irq, &msg);
680		if (status < 0) {
681			destroy_msi_irq(irq);
682			break;
683		}
684
685		write_msi_msg(entry, &msg);
686		attach_msi_entry(entry, irq);
687	}
688	if (i != nvec) {
689		int avail = i - 1;
690		i--;
691		for (; i >= 0; i--) {
692			irq = (entries + i)->vector;
693			msi_free_irq(dev, irq);
694			(entries + i)->vector = 0;
695		}
696		/* If we had some success report the number of irqs
697		 * we succeeded in setting up.
698		 */
699		if (avail <= 0)
700			avail = -EBUSY;
701		return avail;
702	}
703	/* Set MSI-X enabled bits */
704	enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
705
706	return 0;
707}
708
709/**
710 * pci_msi_supported - check whether MSI may be enabled on device
711 * @dev: pointer to the pci_dev data structure of MSI device function
712 *
713 * MSI must be globally enabled and supported by the device and its root
714 * bus. But, the root bus is not easy to find since some architectures
715 * have virtual busses on top of the PCI hierarchy (for instance the
716 * hypertransport bus), while the actual bus where MSI must be supported
717 * is below. So we test the MSI flag on all parent busses and assume
718 * that no quirk will ever set the NO_MSI flag on a non-root bus.
719 **/
720static
721int pci_msi_supported(struct pci_dev * dev)
722{
723	struct pci_bus *bus;
724
725	if (!pci_msi_enable || !dev || dev->no_msi)
726		return -EINVAL;
727
728	/* check MSI flags of all parent busses */
729	for (bus = dev->bus; bus; bus = bus->parent)
730		if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
731			return -EINVAL;
732
733	return 0;
734}
735
736/**
737 * pci_enable_msi - configure device's MSI capability structure
738 * @dev: pointer to the pci_dev data structure of MSI device function
739 *
740 * Setup the MSI capability structure of device function with
741 * a single MSI irq upon its software driver call to request for
742 * MSI mode enabled on its hardware device function. A return of zero
743 * indicates the successful setup of an entry zero with the new MSI
744 * irq or non-zero for otherwise.
745 **/
746int pci_enable_msi(struct pci_dev* dev)
747{
748	int pos, temp, status;
749	u16 control;
750
751	if (pci_msi_supported(dev) < 0)
752		return -EINVAL;
753
754	temp = dev->irq;
755
756	status = msi_init();
757	if (status < 0)
758		return status;
759
760	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
761	if (!pos)
762		return -EINVAL;
763
764	pci_read_config_word(dev, msi_control_reg(pos), &control);
765	if (!is_64bit_address(control) && msi_ops->needs_64bit_address)
766		return -EINVAL;
767
768	WARN_ON(!msi_lookup_irq(dev, PCI_CAP_ID_MSI));
769
770	/* Check whether driver already requested for MSI-X irqs */
771	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
772	if (pos > 0 && !msi_lookup_irq(dev, PCI_CAP_ID_MSIX)) {
773			printk(KERN_INFO "PCI: %s: Can't enable MSI.  "
774			       "Device already has MSI-X irq assigned\n",
775			       pci_name(dev));
776			dev->irq = temp;
777			return -EINVAL;
778	}
779	status = msi_capability_init(dev);
780	return status;
781}
782
783void pci_disable_msi(struct pci_dev* dev)
784{
785	struct msi_desc *entry;
786	int pos, default_irq;
787	u16 control;
788	unsigned long flags;
789
790	if (!pci_msi_enable)
791		return;
792	if (!dev)
793		return;
794
795	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
796	if (!pos)
797		return;
798
799	pci_read_config_word(dev, msi_control_reg(pos), &control);
800	if (!(control & PCI_MSI_FLAGS_ENABLE))
801		return;
802
803	disable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
804
805	spin_lock_irqsave(&msi_lock, flags);
806	entry = msi_desc[dev->irq];
807	if (!entry || !entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI) {
808		spin_unlock_irqrestore(&msi_lock, flags);
809		return;
810	}
811	if (irq_has_action(dev->irq)) {
812		spin_unlock_irqrestore(&msi_lock, flags);
813		printk(KERN_WARNING "PCI: %s: pci_disable_msi() called without "
814		       "free_irq() on MSI irq %d\n",
815		       pci_name(dev), dev->irq);
816		BUG_ON(irq_has_action(dev->irq));
817	} else {
818		default_irq = entry->msi_attrib.default_irq;
819		spin_unlock_irqrestore(&msi_lock, flags);
820		msi_free_irq(dev, dev->irq);
821
822		/* Restore dev->irq to its default pin-assertion irq */
823		dev->irq = default_irq;
824	}
825}
826
827static int msi_free_irq(struct pci_dev* dev, int irq)
828{
829	struct msi_desc *entry;
830	int head, entry_nr, type;
831	void __iomem *base;
832	unsigned long flags;
833
834	msi_ops->teardown(irq);
835
836	spin_lock_irqsave(&msi_lock, flags);
837	entry = msi_desc[irq];
838	if (!entry || entry->dev != dev) {
839		spin_unlock_irqrestore(&msi_lock, flags);
840		return -EINVAL;
841	}
842	type = entry->msi_attrib.type;
843	entry_nr = entry->msi_attrib.entry_nr;
844	head = entry->link.head;
845	base = entry->mask_base;
846	msi_desc[entry->link.head]->link.tail = entry->link.tail;
847	msi_desc[entry->link.tail]->link.head = entry->link.head;
848	entry->dev = NULL;
849	msi_desc[irq] = NULL;
850	spin_unlock_irqrestore(&msi_lock, flags);
851
852	destroy_msi_irq(irq);
853
854	if (type == PCI_CAP_ID_MSIX) {
855		writel(1, base + entry_nr * PCI_MSIX_ENTRY_SIZE +
856			PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
857
858		if (head == irq)
859			iounmap(base);
860	}
861
862	return 0;
863}
864
865/**
866 * pci_enable_msix - configure device's MSI-X capability structure
867 * @dev: pointer to the pci_dev data structure of MSI-X device function
868 * @entries: pointer to an array of MSI-X entries
869 * @nvec: number of MSI-X irqs requested for allocation by device driver
870 *
871 * Setup the MSI-X capability structure of device function with the number
872 * of requested irqs upon its software driver call to request for
873 * MSI-X mode enabled on its hardware device function. A return of zero
874 * indicates the successful configuration of MSI-X capability structure
875 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
876 * Or a return of > 0 indicates that driver request is exceeding the number
877 * of irqs available. Driver should use the returned value to re-send
878 * its request.
879 **/
880int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
881{
882	int status, pos, nr_entries;
883	int i, j, temp;
884	u16 control;
885
886	if (!entries || pci_msi_supported(dev) < 0)
887 		return -EINVAL;
888
889	status = msi_init();
890	if (status < 0)
891		return status;
892
893	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
894	if (!pos)
895 		return -EINVAL;
896
897	pci_read_config_word(dev, msi_control_reg(pos), &control);
898	nr_entries = multi_msix_capable(control);
899	if (nvec > nr_entries)
900		return -EINVAL;
901
902	/* Check for any invalid entries */
903	for (i = 0; i < nvec; i++) {
904		if (entries[i].entry >= nr_entries)
905			return -EINVAL;		/* invalid entry */
906		for (j = i + 1; j < nvec; j++) {
907			if (entries[i].entry == entries[j].entry)
908				return -EINVAL;	/* duplicate entry */
909		}
910	}
911	temp = dev->irq;
912	WARN_ON(!msi_lookup_irq(dev, PCI_CAP_ID_MSIX));
913
914	/* Check whether driver already requested for MSI irq */
915   	if (pci_find_capability(dev, PCI_CAP_ID_MSI) > 0 &&
916		!msi_lookup_irq(dev, PCI_CAP_ID_MSI)) {
917		printk(KERN_INFO "PCI: %s: Can't enable MSI-X.  "
918		       "Device already has an MSI irq assigned\n",
919		       pci_name(dev));
920		dev->irq = temp;
921		return -EINVAL;
922	}
923	status = msix_capability_init(dev, entries, nvec);
924	return status;
925}
926
927void pci_disable_msix(struct pci_dev* dev)
928{
929	int pos, temp;
930	u16 control;
931
932	if (!pci_msi_enable)
933		return;
934	if (!dev)
935		return;
936
937	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
938	if (!pos)
939		return;
940
941	pci_read_config_word(dev, msi_control_reg(pos), &control);
942	if (!(control & PCI_MSIX_FLAGS_ENABLE))
943		return;
944
945	disable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
946
947	temp = dev->irq;
948	if (!msi_lookup_irq(dev, PCI_CAP_ID_MSIX)) {
949		int irq, head, tail = 0, warning = 0;
950		unsigned long flags;
951
952		irq = head = dev->irq;
953		dev->irq = temp;			/* Restore pin IRQ */
954		while (head != tail) {
955			spin_lock_irqsave(&msi_lock, flags);
956			tail = msi_desc[irq]->link.tail;
957			spin_unlock_irqrestore(&msi_lock, flags);
958			if (irq_has_action(irq))
959				warning = 1;
960			else if (irq != head)	/* Release MSI-X irq */
961				msi_free_irq(dev, irq);
962			irq = tail;
963		}
964		msi_free_irq(dev, irq);
965		if (warning) {
966			printk(KERN_WARNING "PCI: %s: pci_disable_msix() called without "
967			       "free_irq() on all MSI-X irqs\n",
968			       pci_name(dev));
969			BUG_ON(warning > 0);
970		}
971	}
972}
973
974/**
975 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
976 * @dev: pointer to the pci_dev data structure of MSI(X) device function
977 *
978 * Being called during hotplug remove, from which the device function
979 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
980 * allocated for this device function, are reclaimed to unused state,
981 * which may be used later on.
982 **/
983void msi_remove_pci_irq_vectors(struct pci_dev* dev)
984{
985	int pos, temp;
986	unsigned long flags;
987
988	if (!pci_msi_enable || !dev)
989 		return;
990
991	temp = dev->irq;		/* Save IOAPIC IRQ */
992	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
993	if (pos > 0 && !msi_lookup_irq(dev, PCI_CAP_ID_MSI)) {
994		if (irq_has_action(dev->irq)) {
995			printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() "
996			       "called without free_irq() on MSI irq %d\n",
997			       pci_name(dev), dev->irq);
998			BUG_ON(irq_has_action(dev->irq));
999		} else /* Release MSI irq assigned to this device */
1000			msi_free_irq(dev, dev->irq);
1001		dev->irq = temp;		/* Restore IOAPIC IRQ */
1002	}
1003	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1004	if (pos > 0 && !msi_lookup_irq(dev, PCI_CAP_ID_MSIX)) {
1005		int irq, head, tail = 0, warning = 0;
1006		void __iomem *base = NULL;
1007
1008		irq = head = dev->irq;
1009		while (head != tail) {
1010			spin_lock_irqsave(&msi_lock, flags);
1011			tail = msi_desc[irq]->link.tail;
1012			base = msi_desc[irq]->mask_base;
1013			spin_unlock_irqrestore(&msi_lock, flags);
1014			if (irq_has_action(irq))
1015				warning = 1;
1016			else if (irq != head) /* Release MSI-X irq */
1017				msi_free_irq(dev, irq);
1018			irq = tail;
1019		}
1020		msi_free_irq(dev, irq);
1021		if (warning) {
1022			iounmap(base);
1023			printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() "
1024			       "called without free_irq() on all MSI-X irqs\n",
1025			       pci_name(dev));
1026			BUG_ON(warning > 0);
1027		}
1028		dev->irq = temp;		/* Restore IOAPIC IRQ */
1029	}
1030}
1031
1032void pci_no_msi(void)
1033{
1034	pci_msi_enable = 0;
1035}
1036
1037EXPORT_SYMBOL(pci_enable_msi);
1038EXPORT_SYMBOL(pci_disable_msi);
1039EXPORT_SYMBOL(pci_enable_msix);
1040EXPORT_SYMBOL(pci_disable_msix);
1041